fix(scout): stop retiring realtime models on a wrong-endpoint 404

server/src/__tests__/db/realtime-restore-v23.test.ts

@@ -0,0 +1,77 @@
+import { describe, it, expect, beforeAll } from 'vitest';
+import { initDb, getDb } from '../../db/index.js';
+import { selectSweepCandidateIds } from '../../services/model-scout.js';
+
+/**
+ * The scout probes Google with generateContent, but realtime models answer only
+ * over bidiGenerateContent — so Google returns a 404 meaning "wrong endpoint",
+ * which the retirement logic read as "removed". Two working Gemini realtime
+ * models were disabled in production as a result.
+ */
+describe('V23 realtime model restore', () => {
+  beforeAll(() => {
+    process.env.ENCRYPTION_KEY = '0'.repeat(64);
+    initDb(':memory:');
+  });
+
+  it('leaves realtime-only models enabled', () => {
+    const db = getDb();
+    const rows = db.prepare(`
+      SELECT m.model_id, m.enabled FROM models m
+      WHERE EXISTS (
+        SELECT 1 FROM model_capabilities c
+        WHERE c.model_db_id = m.id AND c.enabled = 1 AND c.capability = 'realtime_audio'
+      )
+      AND NOT EXISTS (
+        SELECT 1 FROM model_capabilities c2
+        WHERE c2.model_db_id = m.id AND c2.enabled = 1 AND c2.capability IN ('chat', 'vision')
+      )
+    `).all() as { model_id: string; enabled: number }[];
+
+    expect(rows.length, 'catalog should have realtime-only models').toBeGreaterThan(0);
+    for (const row of rows) {
+      expect(row.enabled, `${row.model_id} must not be left retired`).toBe(1);
+    }
+  });
+
+  it('excludes realtime-only models from the chat probe', () => {
+    const db = getDb();
+    const candidates = new Set(selectSweepCandidateIds(db));
+
+    const realtimeOnly = db.prepare(`
+      SELECT m.id, m.model_id FROM models m
+      WHERE EXISTS (
+        SELECT 1 FROM model_capabilities c
+        WHERE c.model_db_id = m.id AND c.enabled = 1 AND c.capability = 'realtime_audio'
+      )
+      AND NOT EXISTS (
+        SELECT 1 FROM model_capabilities c2
+        WHERE c2.model_db_id = m.id AND c2.enabled = 1 AND c2.capability IN ('chat', 'vision')
+      )
+    `).all() as { id: number; model_id: string }[];
+
+    for (const row of realtimeOnly) {
+      expect(
+        candidates.has(row.id),
+        `${row.model_id} cannot answer a generateContent probe and must not be swept`,
+      ).toBe(false);
+    }
+  });
+
+  it('still sweeps ordinary chat models', () => {
+    const db = getDb();
+    const candidates = new Set(selectSweepCandidateIds(db));
+
+    const chatModels = db.prepare(`
+      SELECT m.id, m.model_id FROM models m
+      JOIN model_capabilities c ON c.model_db_id = m.id
+      WHERE m.enabled = 1 AND m.is_free = 1 AND c.capability = 'chat' AND c.enabled = 1
+      LIMIT 5
+    `).all() as { id: number; model_id: string }[];
+
+    expect(chatModels.length).toBeGreaterThan(0);
+    for (const row of chatModels) {
+      expect(candidates.has(row.id), `${row.model_id} should still be swept`).toBe(true);
+    }
+  });
+});

server/src/__tests__/services/scout-auto-retire.test.ts

@@ -105,3 +105,32 @@ describe('recordGoneStreak', () => {
     expect(recordGoneStreak(db, modelId, true)).toBeNull();
   });
 });
+
+/**
+ * Regression: the scout retired two working Gemini realtime models.
+ *
+ * It probes Google with generateContent, but realtime models only answer over
+ * bidiGenerateContent, so Google returns a 404 that means "wrong endpoint",
+ * not "model removed". isGoneMessage matched the 404 and three cycles later
+ * both rows were disabled — while they were working perfectly in the app.
+ */
+describe('wrong-method 404s are not deprecation', () => {
+  const wrongMethod = [
+    'Google API error 404: models/gemini-3.1-flash-live-preview is not found for API version v1beta, or is not supported for generateContent. Call ModelService.ListModels to see the list of available models and their supported methods.',
+    'Google API error 404: models/gemini-2.5-flash-native-audio-preview-12-2025 is not found for API version v1beta, or is not supported for generateContent.',
+    'API error 404: this model is not supported by the completions endpoint',
+  ];
+
+  for (const msg of wrongMethod) {
+    it(`keeps the model when the endpoint is wrong: ${msg.slice(30, 78)}…`, () => {
+      expect(isGoneMessage(msg)).toBe(false);
+    });
+  }
+
+  it('still retires a genuine Google removal', () => {
+    // No "supported methods" clause — the model really is gone.
+    expect(isGoneMessage(
+      'Google API error 404: This model models/gemini-2.5-pro is no longer available to new users',
+    )).toBe(true);
+  });
+});

server/src/db/index.ts

@@ -61,6 +61,7 @@ export async function initDb(dbPath?: string): Promise<Database.Database> {
   seedModelCapabilities(db);
   // Must follow seedModelCapabilities — that is where the image rows are seeded.
   retireDeadCatalogRowsV22(db);
+  restoreRealtimeModelsV23(db);
   flagPaidGoogleModels(db);
   purgeLegacyBazaarlinkDiscoveries(db);
   ensureUnifiedKey(db);
@@ -1803,6 +1804,52 @@ function migrateModelsV21(db: Database.Database) {
   });
   apply();}
 
+/**
+ * V23 (August 2026): undo the realtime models the scout retired by mistake.
+ *
+ * The scout probes Google with generateContent. Realtime/live models are only
+ * reachable over bidiGenerateContent, so Google answers:
+ *
+ *   "models/gemini-3.1-flash-live-preview is not found for API version v1beta,
+ *    or is not supported for generateContent. Call ModelService.ListModels to
+ *    see the list of available models and their supported methods."
+ *
+ * isGoneMessage() read the 404 as removal and the streak logic disabled two
+ * models that were working perfectly. Both the message guard and the candidate
+ * query are fixed; this restores the rows the bug switched off.
+ *
+ * Scoped to rows whose ONLY capabilities are realtime/audio, so it re-enables
+ * exactly the class the probe cannot legitimately judge and never resurrects a
+ * genuinely retired chat model. Runs after seedModelCapabilities because it
+ * reads model_capabilities. Idempotent.
+ */
+function restoreRealtimeModelsV23(db: Database.Database) {
+  db.prepare(`
+    UPDATE models SET enabled = 1
+     WHERE enabled = 0
+       AND EXISTS (
+         SELECT 1 FROM model_capabilities c
+         WHERE c.model_db_id = models.id AND c.enabled = 1
+           AND c.capability IN ('realtime_audio', 'audio')
+       )
+       AND NOT EXISTS (
+         SELECT 1 FROM model_capabilities c2
+         WHERE c2.model_db_id = models.id AND c2.enabled = 1
+           AND c2.capability IN ('chat', 'vision')
+       )
+  `).run();
+
+  // Clear the streak so a stale counter cannot retire them again on the next
+  // probe before the fixed candidate query excludes them.
+  db.prepare(`
+    UPDATE model_availability SET gone_streak = 0, status = 'unknown'
+     WHERE model_db_id IN (
+       SELECT c.model_db_id FROM model_capabilities c
+       WHERE c.enabled = 1 AND c.capability IN ('realtime_audio', 'audio')
+     )
+  `).run();
+}
+
 /**
  * Adds model_availability.gone_streak for scout-driven auto-retirement.
  * Separate from the migrateModelsV* series because it is a schema change on an